本研究使用臺灣電力公司與中央氣象局建置的閃電偵測系統所得觀測資料，再由前人所提出的各種 閃電躍升演算法，提出適用臺灣地區閃電躍升演算法與參數，以期能對臺灣地區豪大雨即時預警提供指 引。 在分析流程上，首先分析臺灣地區閃電觀測系統資料，整體呈現台電偵測系統的雲中閃電(Intra-Cloud lightning, IC)數量明顯高於氣象局。在針對對流胞與測站兩種不同的坐標觀點、不同偵測範圍與總 閃電率門檻進行評估後，本研究採用測站觀點，以半徑10公里範圍的總閃電資料進行閃電躍升分析。 接著，在各種閃電躍升演算法與參數的134種搭配組合中，以測站降雨強度6mm/10mins門檻值，配合 閃電躍升領先時間1小時計算技術得分。根據得分結果，採用台電偵測系統的總閃電資料，以Gatlin (2010)提出之演算法採用10分鐘權重移動平均及Af= 1分鐘，為本研究建議方法（可偵測機率0.42、臨 界成功指數0.26、誤報率則為0.59 )。採用氣象局偵測系統的總閃電資料，則以Schultz etal. (2009)提出 2〇法採用Af = 1分鐘，為本研究建議方法（可偵測機率0.43、臨界成功指數0.33、誤報率則為0.42 )。 上述針對臺灣地區閃電資料分析與閃電躍升演算法選定結果，可作為未來發展臺灣地區豪大雨預警的參 考。

This study is to propose an applicable lightning jump algorithm with the parameters adapted in Taiwan for the heavy rainfall prognosis based on the lightning data of Taiwan Power Company Total Lightning Detection System (TPCLDS) and Central Weather Bureau Lightning Detection System (CWBLDS). Several previous studies have demonstrated that the lightning jump may occur in advance of the severe weather observed by people by few minutes. Thus, it can be indicative of a useful operational guidance for the severe weather potential. In this study, the first part is about the lightning data adoption. Multiple perspectives that determine the optimal conditions of lightning jump algorithm are evaluated and compared, such as the threshold of total flash rate, the distance between lightning position and a station. Then, the algorithms and the parameter, which constitute 134 criteria based on different parameter combinations, was evaluated. The skills of lightning jump algorithms are verified using the contingency table between the lightning jump signal and the precipitation intensity (6mm/10mins). The result shows that the adaptability of lightning jump mechanism based on lightning detections surrounding the station appears more effective than that following a storm cell in terms of the lead time prior to significant rainfall. The radius of the lightning detection away from a station is taken as 10 km, and the total lightning is preferred. The algorithm given by Gatlin (2010) performs the most skillful based on the total lightning data TPCLDS (probability of detection, critical success index, and false alarm rate are about 0.42, 0.26, and 0.59, respectively). And, the algorithm given by Schultz et al. (2009) performs the most skillful based on the total lightning data CWBLDS (probability of detection, critical success index, and false alarm rate are about 0.43, 0.33, and 0.42, respectively).